arXiv subjects

Mingjun Ma

Publications and source records attributed to Mingjun Ma.

8 recordsLinked to original sources

arXiv2026

Distill What the Student Can See: Fisher-Projected On-Policy Distillation for Vision-Language Models

On-policy distillation (OPD) samples trajectories from the current student policy and minimizes token-level divergence between student and teacher next-token distributions at prefixes along those trajectories. This aligns the distillation states with the student's own generation distribution. However, it still assumes that the complete teacher distribution is an appropriate target across student capacities. In vision--language reasoning, teacher corrections can depend on visual distinctions that a compact student cannot represent. Our target-scaling study shows that, as the target approaches the complete teacher distribution, the student realizes less of the prescribed shift and obtains worse downstream performance. We therefore propose \emph{Fisher-Projected On-Policy Distillation} (FP-OPD), which distills only locally realizable teacher corrections. FP-OPD uses continuous visual perturbations to estimate the student's local visual tangent space and projects the centered teacher--student log-probability gap onto this space under the student's Fisher metric. The resulting capacity-aware target is optimized with full-vocabulary reverse KL on student trajectories, retaining the standard OPD framework. In 8B-to-2B distillation, FP-OPD improves all seven evaluated multimodal benchmarks. It raises the average score by 2.77 points over the pretrained student and by 1.60 points over standard OPD. These results demonstrate that locally realizable teacher corrections provide a more effective target for distilling compact vision--language models.

cs.LG↗

arXiv2025

3DTeethSAM: Taming SAM2 for 3D Teeth Segmentation

3D teeth segmentation, involving the localization of tooth instances and their semantic categorization in 3D dental models, is a critical yet challenging task in digital dentistry due to the complexity of real-world dentition. In this paper, we propose 3DTeethSAM, an adaptation of the Segment Anything Model 2 (SAM2) for 3D teeth segmentation. SAM2 is a pretrained foundation model for image and video segmentation, demonstrating a strong backbone in various downstream scenarios. To adapt SAM2 for 3D teeth data, we render images of 3D teeth models from predefined views, apply SAM2 for 2D segmentation, and reconstruct 3D results using 2D-3D projections. Since SAM2's performance depends on input prompts and its initial outputs often have deficiencies, and given its class-agnostic nature, we introduce three light-weight learnable modules: (1) a prompt embedding generator to derive prompt embeddings from image embeddings for accurate mask decoding, (2) a mask refiner to enhance SAM2's initial segmentation results, and (3) a mask classifier to categorize the generated masks. Additionally, we incorporate Deformable Global Attention Plugins (DGAP) into SAM2's image encoder. The DGAP enhances both the segmentation accuracy and the speed of the training process. Our method has been validated on the 3DTeethSeg benchmark, achieving an IoU of 91.90% on high-resolution 3D teeth meshes, establishing a new state-of-the-art in the field.

cs.CV↗

arXiv2025

OMNISEC: LLM-Driven Provenance-based Intrusion Detection via Retrieval-Augmented Behavior Prompting

Recently, Provenance-based Intrusion Detection Systems (PIDSes) have been widely used for endpoint threat analysis. These studies can be broadly categorized into rule-based detection systems and learning-based detection systems. Among these, due to the evolution of attack techniques, rules cannot dynamically model all the characteristics of attackers. As a result, such systems often face false negatives. Learning-based detection systems are further divided into supervised learning and anomaly detection. The scarcity of attack samples hinders the usability and effectiveness of supervised learning-based detection systems in practical applications. Anomaly-based detection systems face a massive false positive problem because they cannot distinguish between changes in normal behavior and real attack behavior. The alert results of detection systems are closely related to the manual labor costs of subsequent security analysts. To reduce manual analysis time, we propose OMNISEC, which applies large language models (LLMs) to anomaly-based intrusion detection systems via retrieval-augmented behavior prompting. OMNISEC can identify abnormal nodes and corresponding abnormal events by constructing suspicious nodes and rare paths. By combining two external knowledge bases, OMNISEC uses Retrieval Augmented Generation (RAG) to enable the LLM to determine whether abnormal behavior is a real attack. Finally, OMNISEC can reconstruct the attack graph and restore the complete attack behavior chain of the attacker's intrusion. Experimental results show that OMNISEC outperforms state-of-the-art methods on public benchmark datasets.

cs.CR↗

arXiv2025

ActMiner: Applying Causality Tracking and Increment Aligning for Graph-based Cyber Threat Hunting

To defend against Advanced Persistent Threats on the endpoint, threat hunting employs security knowledge such as cyber threat intelligence to continuously analyze system audit logs through retrospective scanning, querying, or pattern matching, aiming to uncover attack patterns/graphs that traditional detection methods (e.g., recognition for Point of Interest) fail to capture. However, existing threat hunting systems based on provenance graphs face challenges of high false negatives, high false positives, and low efficiency when confronted with diverse attack tactics and voluminous audit logs. To address these issues, we propose a system called Actminer, which constructs query graphs from descriptive relationships in cyber threat intelligence reports for precise threat hunting (i.e., graph alignment) on provenance graphs. First, we present a heuristic search strategy based on equivalent semantic transfer to reduce false negatives. Second, we establish a filtering mechanism based on causal relationships of attack behaviors to mitigate false positives. Finally, we design a tree structure to incrementally update the alignment results, significantly improving hunting efficiency. Evaluation on the DARPA Engagement dataset demonstrates that compared to the SOTA POIROT, Actminer reduces false positives by 39.1%, eliminates all false negatives, and effectively counters adversarial attacks.

cs.CR↗

arXiv2024

SPARSE: Semantic Tracking and Path Analysis for Attack Investigation in Real-time

As the complexity and destructiveness of Advanced Persistent Threat (APT) increase, there is a growing tendency to identify a series of actions undertaken to achieve the attacker's target, called attack investigation. Currently, analysts construct the provenance graph to perform causality analysis on Point-Of-Interest (POI) event for capturing critical events (related to the attack). However, due to the vast size of the provenance graph and the rarity of critical events, existing attack investigation methods suffer from problems of high false positives, high overhead, and high latency. To this end, we propose SPARSE, an efficient and real-time system for constructing critical component graphs (i.e., consisting of critical events) from streaming logs. Our key observation is 1) Critical events exist in a suspicious semantic graph (SSG) composed of interaction flows between suspicious entities, and 2) Information flows that accomplish attacker's goal exist in the form of paths. Therefore, SPARSE uses a two-stage framework to implement attack investigation (i.e., constructing the SSG and performing path-level contextual analysis). First, SPARSE operates in a state-based mode where events are consumed as streams, allowing easy access to the SSG related to the POI event through semantic transfer rule and storage strategy. Then, SPARSE identifies all suspicious flow paths (SFPs) related to the POI event from the SSG, quantifies the influence of each path to filter irrelevant events. Our evaluation on a real large-scale attack dataset shows that SPARSE can generate a critical component graph (~ 113 edges) in 1.6 seconds, which is 2014 X smaller than the backtracking graph (~ 227,589 edges). SPARSE is 25 X more effective than other state-of-the-art techniques in filtering irrelevant edges.

cs.CR↗

arXiv2023

NeuronsMAE: A Novel Multi-Agent Reinforcement Learning Environment for Cooperative and Competitive Multi-Robot Tasks

Multi-agent reinforcement learning (MARL) has achieved remarkable success in various challenging problems. Meanwhile, more and more benchmarks have emerged and provided some standards to evaluate the algorithms in different fields. On the one hand, the virtual MARL environments lack knowledge of real-world tasks and actuator abilities, and on the other hand, the current task-specified multi-robot platform has poor support for the generality of multi-agent reinforcement learning algorithms and lacks support for transferring from simulation to the real environment. Bridging the gap between the virtual MARL environments and the real multi-robot platform becomes the key to promoting the practicability of MARL algorithms. This paper proposes a novel MARL environment for real multi-robot tasks named NeuronsMAE (Neurons Multi-Agent Environment). This environment supports cooperative and competitive multi-robot tasks and is configured with rich parameter interfaces to study the multi-agent policy transfer from simulation to reality. With this platform, we evaluate various popular MARL algorithms and build a new MARL benchmark for multi-robot tasks. We hope that this platform will facilitate the research and application of MARL algorithms for real robot tasks. Information about the benchmark and the open-source code will be released.

cs.RO↗

arXiv2023

NeuronsGym: A Hybrid Framework and Benchmark for Robot Tasks with Sim2Real Policy Learning

The rise of embodied AI has greatly improved the possibility of general mobile agent systems. At present, many evaluation platforms with rich scenes, high visual fidelity and various application scenarios have been developed. In this paper, we present a hybrid framework named NeuronsGym that can be used for policy learning of robot tasks, covering a simulation platform for training policy, and a physical system for studying sim2real problems. Unlike most current single-task, slow-moving robotic platforms, our framework provides agile physical robots with a wider range of speeds, and can be employed to train robotic navigation and confrontation policies. At the same time, in order to evaluate the safety of robot navigation, we propose a safety-weighted path length (SFPL) to improve the safety evaluation in the current mobile robot navigation. Based on this platform, we build a new benchmark for navigation and confrontation tasks under this platform by comparing the current mainstream sim2real methods, and hold the 2022 IEEE Conference on Games (CoG) RoboMaster sim2real challenge. We release the codes of this framework\footnote{\url{https://github.com/DRL-CASIA/NeuronsGym}} and hope that this platform can promote the development of more flexible and agile general mobile agent algorithms.

cs.RO↗

arXiv2021

ABCP: Automatic Block-wise and Channel-wise Network Pruning via Joint Search

Currently, an increasing number of model pruning methods are proposed to resolve the contradictions between the computer powers required by the deep learning models and the resource-constrained devices. However, most of the traditional rule-based network pruning methods can not reach a sufficient compression ratio with low accuracy loss and are time-consuming as well as laborious. In this paper, we propose Automatic Block-wise and Channel-wise Network Pruning (ABCP) to jointly search the block-wise and channel-wise pruning action with deep reinforcement learning. A joint sample algorithm is proposed to simultaneously generate the pruning choice of each residual block and the channel pruning ratio of each convolutional layer from the discrete and continuous search space respectively. The best pruning action taking both the accuracy and the complexity of the model into account is obtained finally. Compared with the traditional rule-based pruning method, this pipeline saves human labor and achieves a higher compression ratio with lower accuracy loss. Tested on the mobile robot detection dataset, the pruned YOLOv3 model saves 99.5% FLOPs, reduces 99.5% parameters, and achieves 37.3 times speed up with only 2.8% mAP loss. The results of the transfer task on the sim2real detection dataset also show that our pruned model has much better robustness performance.

cs.CV↗